wg1 overview [email protected] deutscher wetterdienst, d-63067 offenbach, germany
DESCRIPTION
current DA method: nudging. WG1 Overview [email protected] Deutscher Wetterdienst, D-63067 Offenbach, Germany. to be developed: PP KENDA for km-scale EPS. PP Sat-Cloud: use of AMSU-A over land use of cloud info from IR-rad. radar reflectivity (precip): latent heat nudging 1DVar. - PowerPoint PPT PresentationTRANSCRIPT
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
WG1 Overview
[email protected] Deutscher Wetterdienst, D-63067 Offenbach, Germany
current DA method: nudgingto be developed: PP KENDA
for km-scale EPS
PP Sat-Cloud: • use of AMSU-A over land• use of cloud info from IR-rad
radar reflectivity (precip): • latent heat nudging• 1DVar
radar radial wind: • for nudging: VAD, SAR, nudg. Vr
ground-based GPS humidity • tomography (profiles)• vertically integrated WV
scatterometer10-m wind
improved use of surface-level obs
LETKFfor HRM
var. soil moisture analysis
snow (cover) PP COLOBOC
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
• DWD: Done: – microphysics change (2006) reduced evaporation below cloud base
ratio RRsurf / RRupper-air and hence RRsurf / RRref increased
(need to) revise definition of reference precipitation reduced overestimation of precipitation during LHN very small impact on forecasts
– bright band detection inside COSMO model
Outlook: – extend use of radar data to foreign radars– revise reference precipitation to account for (min.) radar beam height– better understand how (nature/) model develops convection
(role of environment, (moisture) balance, …)
Use of Radar-derived Surface Precipitation:
Latent Heat NudgingKlaus Stephan (DWD), Daniel Leuenberger (MetCH)
Talk: On the Value of Radar-Derived Rainfall Assimilation on High-Resolution QPF
• MetCH: LHN introduced operationally 20 June 2008, extensive verification done
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Bright Band detection (inside COSMO model)
Quelle: wikipedia
H_zero: height of freezing level in the modelH_radar: height of radar beamRR_RAD: hourly sum of precip. observed by radar
H_zero
H_radarBright Band criteria:
1. H_zero – H_radar [-300;600] RR_RAD(i,j)
2. > 8.5 <RR_RAD>
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview Synop-Regnie Radar g.pts. with BB (≥1x/day)
ASS, LHN, no BB detect. ASS with BB detection ASS without LHN
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Use of Radar-derived Surface PrecipitationVirginia Poli (ARPA-SMR)
Poster: Assimilation of radar derived surface rain rate into the regional COSMO model through a 1D-Var+nudging scheme: analysis of results
• ARPA-SMR: 1DVAR to retrieve T, q –profiles from RR (using linearised parameterisations of large-scale condensation and convection) then nudge T, q –profiles
Model space
Observation space
no
yes
Analysisxa=x
MINIMIZATION?0 Jx
bxJ
0J
Backgroundxb=(Tb, qb, ps
b)
x=(T, q, ps)
Initializationx=xb
0JRR
Jx
0Jx
bJ
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Example of RR assimilation
Very encouraging results!
Shades: Radar observationContours: COSMO-I2 forecasted RR
Control run – Forecast +6 hours
Control run – Forecast +1 hour Experimental run – Forecast +1 hour
Experimental run – Forecast +6 hours
Assimilation of RR is able to dry off precipitation and also to create structures in the right place
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Simple Adjoint Retrieval (SAR) of 3-D Wind VectorJerzy Achimowicz (IMGW) (W.P.1.1.2)
• input data: 3 consecutive scans of 3-d reflectivity and radial velocity at 10’-intervals, interpolated to Cartesian grid (1km x 1km x 500m , 20 levels)
smoothcontm JJJVVWWJ Bob
rrt
rob
tm mmm
2
,212
,21)(
xx
v
‘predicted’ by2100 ,, tob
tt mvvhhmt Fkk 22v
• SAR method: very sensitive to errors in radial velocity input data new software package developed for QC of radar Doppler data
(incl. de-aliasing, interpolation from polar to cartesian coord.)
Doppler radial velocity
wind retrieval
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
IWV derived from observed TZD (with p, T from Synop or COSMO)
Use of Integrated Water Vapour (IWV) from Ground-Based GPSMariella Tomassini, Klaus Stephan, Christoph Schraff (DWD) (W.P. 1.2)
q model
q gps
IWV gps < IWV mod
kqIWV
IWVkq v
obsobsv
modmod
pseudo-obs profile of specific humidity
kw
kpkqkw
satv
max
‘quality weights’ for ( ~ 1 betw. 700 – 800 hPa) : kqobs
v
• IWV from 169 Sta. every 15 min.(verify well with RS92-humidity, except for 12-UTC dry bias of RS92 in summer)
• 1 – 13 June 2007, anticyclonic air-mass convection
• 21-h forecasts from 0, 6, 12, 18 UTC ass cycle
• comparison: ‘CNT’ : like opr (with RS + LHN)‘GPS’ : CNT + GPS‘noRSq : CNT – RS-humidity
• GPS assimilated like radiosonde humidity profiles, but with smaller horizontal influence ( ~120 km → ~ 50 km)
Experiment
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
12 UTC
Analysis00 UTC
06 UTC
Obs18 UTC
CNT NoRSq
GPS
daily cycle of: IWV
CNT 00
GPS 00
CNT 12GPS 12
COSMO-DE too moist
12-UTC RS dries
GPS dries except at 12-UTC
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
RS verification : BIAS (model - obs)
+ 0 h
+ 6 h
CNT GPS NoRSq
00-UTC runs
+ 0 h
+ 6 h
12-UTC runs
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Synop verification
00 UTC Forecast
06 UTC Forecast
12 UTC Forecast
18 UTC Forecast
Correct Cloud Cover Percent : GPS oooo CNT ****
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
hourly mean of precipitation (forecasts compared to radar)
+ + + + +
Obs
CNT
GPS
NoRSq
0.1 mm/h 0.1 mm/h
2.0 mm/h
00 UTC runs 12 UTC runs
2.0 mm/h
reduction of precip by GPS
increase of precip without RS-q
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
radar verification – ETS
0.1 mm/h 0.1 mm/h
1.0 mm/h 1.0 mm/h
00 UTC runs 12 UTC runs
+ + + + +
CNT
GPS
NoRSq
great improvement by GPS
GPS: worsebecause too little
strong precip in early evening
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
• GPS IWV obs from GFZ have good quality
further comparison / assimilation with GPS data from ~ 1000 European stations
(Eumetnet Project E-GVAP) main objects: data selection, extrapolation to 10 m, vertical + horizontal structure functions
• GPS data have shown 12-UTC dry bias of RS92 (in 2007) validate new version of RS92
• GPS data useful for verification of daily cycle of humidity in the model
test future development in data assimilation / physics with these data
• GPS IWV assimilation reduces overestimation of precip at night and has significant positive impact in first 8 hours of 0-UTC forecasts, but tends to suppress strong precip in afternoon
test again, when model physics improve daily cycle of precip, and test in winter
GPS – IWV : Conclusions & Outlook
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Experiment 28 Feb – 9 March 2008 , with QuickScat & ASCAT datawith ASCAT / QuickScatno scatt
pmsl (model – obs)
too low
too strong
gradient
COSMO-EU
9-h forecasts,
valid for
6 March 2008,
9 UTC
Assimilation of Scatterometer 10-m WindHeinz-Werner Bitzer (MetBW), Alexander Cress, Christoph Schraff (DWD) (W.P. 1.5)
(10-m wind nudging with surface pressurecorrection which is in geostrophic balance with 10-m ana. incr.)
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
• aim: replace additional model runs by parameterized regressions to the determine the gradient of the cost function in the variational scheme(absolutely required for GME (long term dry drift), welcome for COSMO model)
Soil Moisture InitialisationMartin Lange, Werner Wergen (DWD) (W.P.1.8.1)
)()()()( 221
221 obs
mmTobs
mmbT
b TTOTTwwBwwJ
errorfcmT
bmobs
mT
mTmTT
mTbana wTTOBOww2
221
211
21
2 ))(()(
Cost function penalizes deviations from observations and initial soil moisture content
Analysed soil moisture depends on T2m forecast error and sensitivity T2m/w
0J
)00:0,(
)00:15,00:12(2
kw
T m
current scheme: by additional model runswith slightly different w(k,0:00)
new scheme: parameterised as a function of predicted latent heat flux at noon
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Deutscher WetterdienstDeutscher WetterdienstDeutscher Wetterdienst
T2m (12 & 15 UTC) : good performance in summer , degredation in winter
Bias T2m on LM1 domain, avg 12:00, 15:00 RMSE T2m on LM1 domain, avg 12:00, 15:00
comparison of parameterised SMA with operational SMA:experiment May – November 2006
no SMA opr. SMAparam. SMA
no SMA opr. SMAparam. SMA
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Deutscher WetterdienstDeutscher WetterdienstDeutscher Wetterdienst
Small change in top layers, higher wetness in bottom layers
soil moisture contentRMS of SMA increments, at layer 4 (9-27cm)( SMA incr. at layer 5 = 3 * (SMA incr. at layer 4) )
opr. SMA : top layerparam. SMA: top layer opr. SMA : bottom layerparam. SMA: bottom layer
opr. SMA param. SMA
• small differences in upper layers (until Nov.)• stronger moistening of lower layers
(further reduces positive T2m bias in summer)
comparison of parameterised SMA with operational SMA:experiment May – November 2006
parameterised SMA : almost zero increments during winter,
starting mid September
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
22
21
1
22 w
w
Tw
w
TT mm
m
nevaporatio
m
drainagenalgravitatio
mm
w
T
w
w
w
T
w
T
2
2
2
1
1
2
2
2
total differential:
sensitivity of T2m to w2
is different in operational and parameterised SMA in winter
parameterised(in winter: near zero
due to inactive plants)
not parameterised,but how does it look like
in the model (i.e. in the operational SMA)
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
8 cm
15 cm
60 cm
90 cm
soil water content: Lindenberg observations
15 cm: reacts after 6 hours
5 – 7 Nov 2006 (2 days)15 Oct 2006 – 1 Jan 2007 (2.5 months)
30 cm: reacts after 4 days
45 cm: reacts after 2 weeks
→ expect model layer 27 – 81 cm to take about 1 week to react
→ expect model layer 9 – 27 cm to take few hours at most to react
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
soil water content: model at Lindenberg
model layer 27 – 81 cm expectedto take about 1 week to react→ ok
model layer 9 – 27 cm expectedto take few hours at most to react→ ok
→ gravitational drainage (sedimentation) appears roughly realistic in COSMO→ soil moisture increments of operational SMA appear reasonable
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
parameterised SMA for COSMO: operational in 2009 (spring (?): simple version, autumn: with gravitational drainage)
drainagenalgravitatio
m
w
w
w
T
2
1
1
2
Outlook
parameterise also
can be derived analytically from Richards eq. used in COSMO (TERRA)
parameterisation already exists in current version of param. SMA
parameterised SMA for GME: full experiment started, operational in spring 2009
include RH2m as additional obs (param. implemented, increments reasonable in first case)
possible further extensions:• Analyse the top 5 soil layers separately instead of 2 aggregated layers (DWD).• Inclusion of precipitation analysis when good product is available (Suisse).• Improvement of model error statistics (Italy).
Note: SMA parameterisation needs some maintenance to account for future changes in the parameterisation of surface fluxes e.g. modification of root water uptake
cheap, efficient
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Thank you for your attention
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Advantages of NetCDF:• widely used and portable • a variety of software exists to plot, analyse and evaluate the data.
DWD plans: envisaged set-up observation formats, pre- and post-processingDWD plans: envisaged set-up observation formats, pre- and post-processing
can keep AOF asalternative data input
as long as needed
DWD switches to NetCDF on 17 Sept. 2008thereafter, DWD will no longer support AOF interface
~ 1 by 1 convertersimple + portableapplicable to WMO or non-WMO BUFR
standard WMO templates,i.e. unique descriptors + dimensions of elements + code tablesunique BUFR format for each obs type
NetCDF 2
ODBODB monitoring
NetCDFobs
3DVar NetCDFfeedbac
kCOSMOmodel
verificationNWPsection
any kind of
BUFR
bufr 2 wmo_bu
fr
WMOBUFR
bufr2netcdf
IT section
SKY /archive
Under discussion at DWD
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
analysis operational new T2m diagnostics
Deutscher Wetterdienst
COSMO-EU 20070427 00:00 +15 hours
New T2m diagnostics affects the whole PBL through SMA
Bias T2m, C-EU on LM1-domain, avg12:00, 15:00 Accumulated soil moisture increments
Rmse T2m, C-EU on LM1-domain, avg12:00, 15:00
10 m
2250 m
Dew point temperature Germany
both runs done with operational version
of SMA
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
mean skill scores over 32 forecast (00 and 12 UTC) AUGUST 2006threshold 0.1 mm/h
ETSFBI
ASS FORECAST ASS FORECAST
LHN and prognostic precipitation
shows impact of LHN refinements in 2005 / 06 (reference precip / LHN restricted to ‘cloudy layers’ / grid point search / limits)
Stephan, K., S. Klink, C. Schraff, 2008: Assimilation of radar-derived rain rates into the convective-scale model COSMO-DE at DWD. Q. J. R. Meteorol. Soc., 134, 1315 – 1326.
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
New PP: Km-scale Ensemble-based Data Assimilation (KENDA)
Discussion with input from Chris Snyder 18 Sept 2007 on EnKFDiscussion with input from Chris Snyder 18 Sept 2007 on EnKF
– no new obstacles seen for the EnKF
– to get a system to evaluate, need 2 people (with good background) for 2 years
– do EnKF first without radar data (quality control problems), gain experiences, detect bugs / flaws in the scheme, later include radar data
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Assimilation of Scatterometer Wind
29 Feb 08, 0 UTC
COSMO-EU ana with ASCAT/QuickScatCOSMO-EU ana , no scatt
ECMWF analysis 29 Feb 08ASCAT 28 Feb 08, 21 UTC ± 1.5h
984 hPamax. 30 kn
~15 m/s
10-m wind [m/s]
COSMO General Meeting, Cracow, 15 – 19 Sept. 2008Overview on Data Assimilation [email protected]
WG1 Overview
Variational assimilation
Model space
Observation space
no
yes
Analysisxa=x
MINIMIZATION?0 Jx
bxJ
0J
Backgroundxb=(Tb, qb, ps
b)
x=(T, q, ps)
Initializationx=xb
0JRR
Jx
0Jx
bJ
Convert observations (Rain Rates - RR) in profiles of temperature and humidity and nudge them as “pseudo”-observations.